110+ queued PRs cleared in 30 hours

Problem

Reeve is an AI commerce-growth platform, and its MVP was waiting on a queue of more than 110 pull requests across 4 repositories. I was Reeve’s release gate, so clearing that queue was my job — without letting speed lower the bar for what reaches production.

A queue that size fights itself. Every merge moves the base under the pull requests behind it.

Approach

I split the work in two: parallel preparation, serial admission.

1. 01Queue110+ PRs in 4 repos
2. 02Agentsa worktree each
3. 03Gatetests on every merge
4. 04Merged110+ in ~30 h

1. 01More than 110 pull requests waited across 4 repositories, and every merge moved the base under the rest.
2. 02Claude Code agents, each in its own git worktree, prepared the pull requests and stacked dependent changes into chains.
3. 03One at a time, every pull request passed a serial merge gate that ran the tests on the merge result.
4. 04The whole queue was cleared in about 30 hours. The agents supplied the speed; the gate kept the bar.

Step 1 of 4

More than 110 pull requests waited across 4 repositories, and every merge moved the base under the rest.

Preparation ran in parallel. Claude Code agents each worked in an isolated git worktree, so they never stepped on each other. They assembled dependent changes into stacked PR chains that could land in order.

Admission stayed serial. Every pull request passed a merge gate before it merged: the tests ran on every merge result, on a sharded Docker replica of CI. Along the way I fixed security findings as well.

Result

The whole queue, more than 110 pull requests, was cleared in about 30 hours. The agents supplied the throughput; the serial gate supplied the safety.

It is the same role I held day to day as Reeve’s release gate: 1,900+ pull requests merged, mostly teammates’.